US2025315711A1PendingUtilityA1

Quantum machine learning method for multi-class classification

Assignee: UNIV KOREA RES & BUS FOUNDPriority: Oct 6, 2023Filed: Dec 19, 2024Published: Oct 9, 2025
Est. expiryOct 6, 2043(~17.2 yrs left)· nominal 20-yr term from priority
G06N 10/60B82Y 10/00G06N 3/047G06N 3/0464G06N 10/20G06N 10/40
65
PatentIndex Score
0
Cited by
0
References
0
Claims

Abstract

The present invention relates to a quantum machine learning method for multi-class classification, and the method comprises the steps of: applying a Quantum Convolution Neural Network (QCNN) quantum circuit to input data having q qubits, and outputting a feature vector based on Pauli-Z measurement; and applying a Quantum Neural Network (QNN) quantum circuit to the feature vector, and outputting a multi-class prediction vector with scalability increased compared to q qubits based on basis measurement.

Claims

exact text as granted — not AI-modified
1 . A quantum machine learning method, the method comprising:
 applying a Quantum Convolution Neural Network (QCNN) quantum circuit configured to input data having q qubits, and outputting a feature vector based on Pauli-Z measurement; and   applying a Quantum Neural Network (QNN) quantum circuit to the feature vector, and outputting a multi-class prediction vector with greater scalability than q qubits based on basis measurement.   
     
     
         2 . The method of  claim 1 , wherein the outputting a multi-class prediction vector inputs the feature vector into the QNN quantum circuit, and outputs a 2 q -dimensional observation value as a probability measurement for the multi-class prediction vector based on the basis measurement. 
     
     
         3 . The method of  claim 1 , wherein the outputting a feature vector comprises:
 receiving an image as the input data through a plurality of input channels as an input, and outputting an initial feature vector based on the Pauli-Z measurement; and   receiving the initial feature vector as an input, and outputting a feature vector having q qubits for each input channel based on the Pauli-Z measurement.   
     
     
         4 . The method of  claim 3 , further comprising: including one or more QCNN layers, and inputting an initial feature vector output by each QCNN layer based on the Pauli-Z measurement into a next QCNN layer. 
     
     
         5 . The method of  claim 1 , wherein a probability amplitude regularizer is used as a loss function in learning the QNN quantum circuit, and a probability amplitude normalizer L PAR  is configured to remove probability values for classes that are not used in learning the QNN quantum circuit, based on the following equation: 
       
         
           
             
               
                 
                   
                     ℒ 
                     
                       P 
                       ⁢ 
                       A 
                       ⁢ 
                       R 
                     
                   
                   ( 
                   
                     Θ 
                     ; 
                     X 
                   
                   ) 
                 
                 = 
                 
                   
                     - 
                     
                       
                         ∑ 
                           
                       
                       
                         
                           n 
                           ′ 
                         
                         > 
                         
                           
                             ❘ 
                             "\[LeftBracketingBar]" 
                           
                           y 
                           
                             ❘ 
                             "\[RightBracketingBar]" 
                           
                         
                       
                       
                         2 
                         q 
                       
                     
                   
                   ⁢ 
                   
                     log 
                     ⁡ 
                     ( 
                     
                       1 
                       - 
                       
                         p 
                         
                           n 
                           ′ 
                         
                       
                     
                     ) 
                   
                 
               
               , 
             
           
         
         q≥┌log 2 (|y|)┐, |y| is the number of classes of a task to be classified, 2 q  is an output of QNN, Θ is trainable parameters, X is extracted features, and Pn′ is an n′-th projection matrix. 
       
     
     
         6 . The method of  claim 5 , wherein a binary cross-entropy loss and a probability amplitude regularizer are used as loss functions in learning the QCNN quantum circuit and the QNN quantum circuit, and the binary cross-entropy loss L BCE  is defined as shown in the following equation: 
       
         
           
             
               
                 
                   
                     ℒ 
                     BCE 
                   
                   ( 
                   
                     Θ 
                     ; 
                     X 
                   
                   ) 
                 
                 = 
                 
                   - 
                   
                     
                       
                         ∑ 
                           
                       
                       
                         n 
                         = 
                         1 
                       
                       
                         | 
                         y 
                         | 
                       
                     
                     [ 
                     
                       
                         
                           y 
                           c 
                         
                         ⁢ 
                         log 
                         ⁢ 
                         
                           p 
                           c 
                         
                       
                       + 
                       
                         
                           ( 
                           
                             1 
                             - 
                             
                               y 
                               c 
                             
                           
                           ) 
                         
                         ⁢ 
                         
                           log 
                           ⁡ 
                           ( 
                           
                             1 
                             - 
                             
                               p 
                               c 
                             
                           
                           ) 
                         
                       
                     
                     ] 
                   
                 
               
               , 
             
           
         
         y c  and p c  
 denote the one-hot encoded class label and the predicted probability, respectively. 
 
       
     
     
         7 . The method of  claim 6 , wherein a train loss function for one-step single update is defined as shown in the following equation: 
       
         
           
             
               
                 
                   ℒ 
                   ⁡ 
                   ( 
                   
                     Θ 
                     ; 
                     ζ 
                   
                   ) 
                 
                 = 
                 
                   
                     1 
                     
                       
                         ❘ 
                         "\[LeftBracketingBar]" 
                       
                       ζ 
                       
                         ❘ 
                         "\[RightBracketingBar]" 
                       
                     
                   
                   ⁢ 
                   
                     
                       
                         ∑ 
                           
                       
                       
                         
                           ( 
                           
                             X 
                             , 
                             y 
                           
                           ) 
                         
                         ∈ 
                         ζ 
                       
                     
                     [ 
                     
                       
                         
                           ℒ 
                           BCE 
                         
                         ( 
                         
                           Θ 
                           ; 
                           X 
                         
                         ) 
                       
                       + 
                       
                         
                           ℒ 
                           PAR 
                         
                         ( 
                         
                           Θ 
                           ; 
                           X 
                         
                         ) 
                       
                     
                     ] 
                   
                 
               
               , 
             
           
         
       
       wherein ζ denotes a set of sampled mini-batch, consisting of the data X and the label y.

Join the waitlist — get patent alerts

Track US2025315711A1 — get alerts on status changes and closely related new filings.

We store only your email — no account needed. See our privacy policy.